Gemini 3.5 Flash
2026-05-19 · Publisher : Google · Domain : Artificial Intelligence
Google introduces the Gemini 3.5 family at I/O with 3.5 Flash, its strongest model for agents and coding, available immediately, ahead of Gemini 3.1 Pro on Terminal-Bench 2.1 and MCP Atlas.
What changes
- Surpasses Gemini 3.1 Pro on agentic and coding benchmarks: Terminal-Bench 2.1, GDPval-AA, MCP Atlas.
- About four times faster than other frontier models in output tokens per second, according to Google.
- Default model of the Gemini app and AI Mode; engine of the Gemini Spark personal agent.
- Works with the Antigravity harness and subagents for long tasks.
- Gemini 3.5 Pro announced as tested internally, with release planned the following month (not published as of September 29, 2026, see notes).
### Specifications - Context: up to 1 million tokens; output up to 64,000 tokens. - Modalities: text, images, audio, video as input; text as output. - Availability: GA in Antigravity, Gemini API (AI Studio), Android Studio, Gemini Enterprise Agent Platform, Gemini app, AI Mode. - API identifier: gemini-3.5-flash. Knowledge cutoff: not extracted from the page. ### Pricing | Item | Standard (USD / 1M tokens) | Batch | |---|---|---| | Input | 1.50 | 0.75 | | Output | 9.00 | 4.50 | Input cache: 0.15 USD. Pricing recorded from the current pricing page, consistent with the comparison in the 3.6 Flash post. ### Benchmarks | Benchmark (scores published by the vendor) | 3.5 Flash | 3.1 Pro | GPT-5.5 | |---|---|---|---| | Terminal-Bench 2.1 | 76.2% | 70.3% | 78.2% | | SWE-Bench Pro | 55.1% | 54.2% | 58.6% | | MCP Atlas | 83.6% | 78.2% | 75.3% | | OSWorld-Verified | 78.4% | 76.2% | 78.7% | | GDPval-AA (Elo) | 1656 | 1314 | 1769 | | CharXiv Reasoning | 84.2% | 83.3% | 84.1% | | Humanity's Last Exam | 40.2% | 44.4% | 41.4% | | ARC-AGI-2 | 72.1% | 77.1% | 84.6% | ### Key takeaways A Flash model that rivals the previous generation's Pro models on agent and coding tasks, at a lower cost.
Links
Tags : LLM, Google, Gemini, Flash, Agents